regional / local data beyond supply -...

SCORUS-Conference • Shanghai Oct. 2008 1

Regional / Local Data

beyond Supply

By Hans Menge

SCORUS-Conference • Shanghai Oct. 2008 2 2

Target of this presentation

How to provide

regional or local data

in case a

National System of Statistics

does not supply

the variables requested for

regional or local level "xyz“


Reasons for „no supply“ YOUNG administration or young system of statistics,

not yet so well developed

OLD statistical framework, downgraded to reduce the burden

on respondents, increased use of administrative data

My hypothesis:

There is more IN the national systems of statistics

(as possible source of regional information)

than is usually pulled OUT

Compared with the expenses for implementing and operating

a new survey or for enlarging a sample survey,

it is rather inexpensive to implement a system of calculation

(even a comprehensive and sophisticated system)

and execute all the data analysis needed


Methodological approaches to

Regional Disaggregation (R.D.), specified by

3 Examples

1 Regional Accounts

All Regional Account variables are generated by

R.D.

2 Urban Audit

97 of 333 variables for the German Urban Audit

to be generated more or less by R.D.

3 Labour Force Survey in Turkey

Upgrade the Turkish Statistical System

by generating regional data (NUTS-3)

with the aid of R.D., using NUTS-2 data


REGIONAL DISAGGREGATION

in Regional Accounts

Top down: Preset national benchmark figure must be scored by the sum of all regional results

Bottom up: The sum of the available (auxiliary) regional variable should score the national benchmark figure !

Requested: most appropriate (= best correlated) regional variable to suit the national variable (benchmark)

Problem of Adequacy Example: „Gainfully Employed” as key variable to disaggregate some part of GDP ? …

Does that fit ? You have to face large regional differences of … • full-time.../ half-time... / marginal employment • contracted hours • qualification and monetary compen- sation


Regions:

Auxiliary variable

[Pers.]

National

Benchmark

[Pers.]

var. A 1

Regional

Proportion

Sum of

Regional

Variable

which differs

from bench-

mark variable

Difference, to be distributed proportionally

(accord. to Var. A2) to all regional units

Results

Regional figures

var. A 2 var. A 1

Region 11

Region 12

Region 13

Region 13

( … )

1) Variables of same dimension, e.g.: national: A1 = „Gainfully Employed in total“ regional: A2 = „Registered Employees in total"


2nd Example:

Urban Audit (German data) 333 variables required in total

97 of which had to be „estimated“ for German U.A.

because: lack of census data

No statistical source with results at regional level available for these

variables

Only „Estimation“ ?

No ! – because:

Validity of results is comparable to those of Reg. Accounts

How ?

• Use of upgraded system of calculation,

• based on Regional Accounts methods, but

• more sophisticated calculations (to fit special requirements)

Special Problems … (see next slide)


Special Problems • Special Solutions

Non Totable variables, e.g.:

„%“, „sqm / household“, „EUR/hh"

Complementary variables

(„Sex“, „modal split of commuters“)

Nested variables, e.g.:

„A“, „B“ and „total of A living in B“

Classified variables (by data supply) However, required: „Quinitil“ etc...

Unsuitable Reference Time Required: 2004,

supplied: 2002.

Different Regional levels (and code systems): NUTS-3 or

S of NUTS-3 or part of NUTS-3

Refining Methods: 2 different paths

acc. to types of numerator / denominator

Iterative Proportional Fitting (IPF)

to ensure consistent results

Special Trick to ensure consistent results

Satellit Calculation Systems for diverse extraction and conversion of the demanded variables

Different Techniques of Regional

Disaggregation & Aggregation

Projection of basic Suryey 2002 by appropiate key variables on regional level


Are there Alternatives to avoid Deduction of data by Regional Disaggregation ?

Alternatives:

Upgrade Sample Size ?

More Regional Detail from the Sample Survey

itself ?

Effort to Steady /

Stabilize annual data base ?

(to avoid mavericks

in time series)

Answer:

Not feasible because of high costs

involved

No valid results on lower level Even Supplied Regional data provide:

a) plenty of blocked data

b) some problematic results

Apply Sample Survey of each year ! … and Smooth time series

by „EMA“ (Exponential Moving Average)

= „Temporal Aggregation“ Why EMA ?

Well known simple „moving average“ has a problem …

– at the current edge and a problem of

– unsteadiness through former periods


Why not enlarge the calculations?

To execute the system of calculations for • 85 Core Cities and LCA on behalf of Urban Audit – or for all … • 439 regional units of level NUTS-3 means only a slight increase in effort & expense !

Thus:

It would be reasonable to do it for all regions ! – In oder to generate general improvement of regional data base !

! 85 Core

Cities

(Urban Audit) 439 regional units

(level NUTS-3

+30%

effort

means

+ 500%

data !

Total Effort to provide

439 regional units

(level NUTS-3

439 Regional

Units

(level NUTS-3)


3rd Example: Turkey: Statistics to be upgraded (here: to be refined)

Task:

Break down

Labour Force

Survey data

(„LFS“ 2004-2007)

from NUTS-2

to NUTS-3

Problem:

Missing Key variables at level NUTS-3 (which were up to date !), as needed for standard disaggregation methods

Supplied only: • Census 2000 [too old !], • Current population projection [no adequate key variable to disaggregate „Labour“ ]

What to do ?


Way out: Combined Methodical Approach

A „Structural Method“ Use strongly discriminating Structural Variables within LFS: • Urban / Rural • Male / Female

Adaptation & Projection of Census 2000 figures

1) Compare Census 2000 with old LFS 2000, then Adapt Census figures

2) „Disaggregate“ Census 2000 [NUTS-2 to NUTS-3] by „Structural Method“ and Compare with real NUTS-3 results, then derive factors for EACH regional unit (to adjust results of Method A)

C Use register data (for Employed)

as soon as their validity proves to be sufficient for this purpose. Register is being installed in Turkey, but not yet completed. (Problem: Different stage of completeness)

D Efforts to ensure valid data • Use all helpfull techniques

(IPF, Exponent. Moving Average, etc.)

• Lots of data analysis at EACH

step of the Calculation system to ensure consistency to distinguish between regional particularities and mavericks and to identify simple mistakes within your calculation system

Means: Calculate on lower level than requested


Conclusions

Data IN >> Data OUT

There is more possible Information IN some

national Survey Data than usually pulled OUT

How ? – by special Techniques

There are simple Methods & assistant Techniques

to ensure valid Regional Data

These have to be

• adapted to the challenges of each specific data supply & request

and

• joined to a comprehensive & circumspect system of calculations &

plausibility control

Chance to put into practice ?

Efforts to generate and apply these procedures are rather small,

compared with the expenses for (enlarging) a survey


Thank you for your attention

Hans Menge

53175 Bonn / Germany

email: [email protected]

Tel. +49 228 317438

… and thank you, Berthold, for the Presentation

… Thank you, • Klaus Trutzel / Germany for supervising German Urban Audit methods and … • Didem Sezer / Turkey for initiating & attending further development of Methods

mailto:[email protected]

regional / local data beyond supply -...

Documents